feat(authorization-obligations): export obligation shapes through the existing store

Task 1.3 (Req 2.1, 2.2, 2.4).

- seed.pair_lessons derives obligation shapes for absence-mechanism rows
  and falls back to the sink shape when no obligation is learned, so the
  flow store is unchanged; skip reasons name the accepted dischargers.
- append_from_pair accepts both shape families (guard carries the
  discharger kind); search_tree ignores obligation rows;
  obligation_mechanisms selects them.
- Measured on vibe-py: the two vampi pairs seed protected_read /
  identity_constraint / authenticated_context.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_01T7XD7x5KonMdmGRkp63EQG

src/openultrasast/semantic/mechanisms.py

@@ -14,6 +14,7 @@ from pathlib import Path
 from ..complexity.map import Hotspot
 from ..findings import StaticFinding
 from ..index import CodeChunk, EmbeddingClient, VectorIndex, VectorRecord, write_vector_index
+from .obligations.shapes import ObligationShape
 from .overlay import OverlayRecord
 from .variants import Shape
 
@@ -96,14 +97,14 @@ def append_mechanism(
 _TIER_RANK = {"seeded": 1, "reviewed": 2}
 
 
-def corpus_mechanism_id(shape: Shape) -> str:
+def corpus_mechanism_id(shape: Shape | ObligationShape) -> str:
     """Deterministic id from the shape key, so the same shape gets the same record on every machine (lever-addressable)."""
     return "corpus:" + hashlib.sha1(shape.key().encode()).hexdigest()[:16]
 
 
 def append_from_pair(
     store: MechanismStore,
-    shape: Shape,
+    shape: Shape | ObligationShape,
     *,
     summary: str,
     cwe: str,
@@ -120,7 +121,15 @@ def append_from_pair(
     if existing is not None and _TIER_RANK.get(existing.review_tier, 0) > _TIER_RANK.get(tier, 0):
         tier = existing.review_tier  # a shape keeps the strongest review it ever had
     inherited = tuple(tag for tag in (existing.tags if existing is not None else ()) if tag.startswith(("provenance:", "tier:")))
-    positions = ", ".join(f"arg{p} ({k})" for p, k in zip(shape.source_positions, shape.source_kinds, strict=True))
+    if isinstance(shape, Shape):
+        positions = ", ".join(f"arg{p} ({k})" for p, k in zip(shape.source_positions, shape.source_kinds, strict=True))
+        exploitable = f"{positions} reaches {shape.sink_name}/{shape.arity}; the fix added {shape.guard}"
+        keywords: tuple[str, ...] = (shape.sink_name, shape.guard, *shape.source_kinds)
+        guard = shape.guard
+    else:  # obligation shape: the discharger the fix carries is the 'guard' the reports show
+        exploitable = f"{shape.operation_kind} reached without {shape.discharger_kind}; the fix binds {shape.provenance}"
+        keywords = (shape.operation_kind, shape.discharger_kind, shape.provenance, shape.resource_class)
+        guard = shape.discharger_kind
     record = Mechanism(
         id=mechanism_id,
         summary=summary,
@@ -129,18 +138,18 @@ def append_from_pair(
         tags=(
             shape.mechanism,
             f"mechanism:{shape.mechanism}",
-            f"guard:{shape.guard}",
+            f"guard:{guard}",
             *sorted({*inherited, f"provenance:{provenance}", f"tier:{tier}"}),
         ),
-        keywords=(shape.sink_name, shape.guard, *shape.source_kinds),
-        what_made_it_exploitable=f"{positions} reaches {shape.sink_name}/{shape.arity}; the fix added {shape.guard}",
+        keywords=keywords,
+        what_made_it_exploitable=exploitable,
         source_finding_id="",
         source_repo="",
         origin="corpus",
         review_tier=tier,
         pairs=pairs,
         shape=shape.to_dict(),
-        guard=shape.guard,
+        guard=guard,
     )
     store.append(record)
     return record

src/openultrasast/semantic/seed.py

@@ -16,6 +16,8 @@ from ..preprocess import LANGUAGE_BY_EXTENSION
 from .facts import FactLoadError, SemanticFacts, load_facts
 from .ir import FileIR, parse_file
 from .mechanisms import MechanismStore, append_from_pair
+from .obligations.facts import ObligationFacts, load_obligation_facts
+from .obligations.shapes import ABSENCE_MECHANISMS, ObligationShape, derive_obligation, explain_skip
 from .variants import Shape, derive_shape
 
 LANGUAGE_UNSUPPORTED = "variants_language_unsupported"
@@ -85,7 +87,7 @@ def export_mechanisms(cases: Sequence[PairCase], store: MechanismStore, *, facts
         if lessons:
             seeded.append(case.name)
         else:
-            skipped.append((case.name, "no_labeled_sink_call_site"))
+            skipped.append((case.name, _skip_detail(case, vuln_ir, fixed_ir, vuln_text, fixed_text, loaded)))
     return ExportReport(
         seeded=len(seeded),
         records=len(record_ids),
@@ -97,9 +99,9 @@ def export_mechanisms(cases: Sequence[PairCase], store: MechanismStore, *, facts
 
 @dataclass(frozen=True)
 class Lesson:
-    """One shape a trusted pair teaches, with the CWE and a text-free summary."""
+    """One shape a trusted pair teaches, with the CWE and a text-free summary (a sink shape or an obligation shape)."""
 
-    shape: Shape
+    shape: Shape | ObligationShape
     cwe: str
     summary: str
 
@@ -125,6 +127,27 @@ def pair_lessons(
     for row in case.expected:
         if not row.function or not row.mechanism:
             continue
+        obligation_kind = getattr(row, "obligation", None)
+        if obligation_kind or row.mechanism in ABSENCE_MECHANISMS:
+            obligation = derive_obligation(
+                vuln_ir,
+                fixed_ir,
+                function=row.function,
+                obligation=obligation_kind,
+                mechanism=row.mechanism,
+                facts=_obligation_facts(),
+                flow_facts=facts,
+                vuln_text=vuln_text,
+                fixed_text=fixed_text,
+            )
+            if obligation is not None:
+                summary = (
+                    f"{obligation.mechanism}: {obligation.operation_kind} without {obligation.discharger_kind} ({row.cwe}); "
+                    f"fix binds {obligation.provenance}"
+                )
+                lessons.append(Lesson(shape=obligation, cwe=row.cwe, summary=summary))
+                continue
+            # an absence row that teaches no obligation may still teach a sink shape (a weak literal at a sink, say)
         shape = derive_shape(
             vuln_ir,
             fixed_ir,
@@ -143,6 +166,32 @@ def pair_lessons(
     return lessons
 
 
+def _skip_detail(case: PairCase, vuln_ir: FileIR, fixed_ir: FileIR, vuln_text: str, fixed_text: str, flow_facts: SemanticFacts) -> str:
+    """Why a parsing, labeled pair taught nothing: the obligation derivation's reason for absence rows, else the sink reason."""
+    for row in case.expected:
+        if row.function and row.mechanism and (getattr(row, "obligation", None) or row.mechanism in ABSENCE_MECHANISMS):
+            return explain_skip(
+                vuln_ir,
+                fixed_ir,
+                function=row.function,
+                vuln_text=vuln_text,
+                fixed_text=fixed_text,
+                obligation=getattr(row, "obligation", None),
+                facts=_obligation_facts(),
+                flow_facts=flow_facts,
+            )
+    return "no_labeled_sink_call_site"
+
+
+_OBLIGATION_FACTS: list[ObligationFacts] = []
+
+
+def _obligation_facts() -> ObligationFacts:
+    if not _OBLIGATION_FACTS:
+        _OBLIGATION_FACTS.append(load_obligation_facts())
+    return _OBLIGATION_FACTS[0]
+
+
 def _skip_reason(case: PairCase) -> str | None:
     if case.review_tier not in GATING_TIERS:
         return f"tier:{case.review_tier}"

src/openultrasast/semantic/variant_search.py

@@ -38,7 +38,8 @@ def search_tree(
     max_mechanisms: int,
 ) -> SearchResult:
     """Variant search over the parsed files of a tree with the store's corpus shapes (bounded by ``max_mechanisms``)."""
-    records = corpus_mechanisms(store.load())[: max(max_mechanisms, 0)]
+    records = [record for record in corpus_mechanisms(store.load()) if (record.shape or {}).get("family", "sink") == "sink"]
+    records = records[: max(max_mechanisms, 0)]
     shapes: list[Shape] = []
     ids: dict[str, str] = {}
     degradations: list[dict[str, object]] = []